AeonLogic

A recursive self-healing multi-agent engine that generates, critiques, repairs, remembers, and reports.

Comment

Inspiration

What it does

How we built it

Inspiration

AeonLogic was inspired by a simple problem: AI tools often generate code once, but real software work needs review, repair, validation, memory, and proof. For the Qwen Cloud hackathon, I wanted to build an agent system that behaves more like a disciplined engineering team than a single chatbot.

What it does

AeonLogic is a recursive self-healing multi-agent engine powered by Qwen Cloud. It takes a software/security task, generates an artifact, critiques it for issues, repairs the detected problems, validates the result, writes lessons into hybrid memory, and produces a final downloadable report.

The Streamlit Command Center shows the complete workflow: Generate → Critique → Repair → Validate → Memory → Final Report.

How I built it

I built the app with Python and Streamlit, with a modular AeonLogic engine behind the dashboard. The system includes a Qwen Cloud-ready client, safe mock fallback mode, critic findings, artifact preview, memory evidence, final verdict reporting, and automated tests with Pytest.

The public demo runs in safe MOCK_MODEL_MODE for reproducible judging and to avoid exposing API keys. The repository includes real-mode Qwen Cloud support through secure environment variables: QWEN_API_KEY, QWEN_BASE_URL, and QWEN_MODEL.

Qwen Cloud / Alibaba Cloud usage

AeonLogic is designed to use Qwen Cloud through an OpenAI-compatible API client. The Qwen client is implemented in: src/aeonlogic/models/qwen_client.py

The app supports qwen-plus as the main model for high-risk/security tasks. API keys are never committed to GitHub. Proof documentation is included in: docs/ALIBABA_CLOUD_PROOF.md docs/ARCHITECTURE.md

Challenges

The biggest challenge was making the project stable for demos while still being real-mode ready for Qwen Cloud. I also had to make the UI clearly show proof of the agent pipeline, critic results, repair output, memory evidence, and final status in a way that judges can understand quickly.

Accomplishments

I am proud that AeonLogic demonstrates a full self-healing AI workflow instead of only showing a simple prompt-response demo. It includes a working Streamlit live demo, Devpost-ready screenshots, a demo video, Qwen Cloud integration code, architecture documentation, and automated tests.

What I learned

I learned how important it is to design AI systems with safety, reproducibility, and explainability. A strong AI agent should not only produce an answer; it should show what it found, how it repaired the issue, what it remembered, and why the final output is approved.

What's next

Next, I want to connect AeonLogic to a real Qwen Cloud API key in production mode, add more agent roles, expand the memory layer, support larger project repair workflows, and deploy an enterprise-ready version for secure software engineering automation.

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for AeonLogic

Built With

  • alibaba-cloud-model-studio
  • chromadb
  • neo4j-ready-memory
  • pytest
  • python
  • qwen-cloud
  • streamlit
Share this project:

Updates